sign in sign up
<<
Home , Palatal consonant, Sonority hierarchy, Voicelessness

41 Studies in African Linguistics Volume 7, Number 1, March 1976

SEGMENT SEQUENCES AND SEGMENTAL

Herbert F. . Stahlke Georgia State University

1. Segmental Discreteness One of the most venerable and venerated postulates in the history of linguistics is the assertion that the continuum is capable of non­ arbitrary division into a linearly ordered string of units, which, de­ pending upon the topic under investigation, can be anything from phonetic segments to whole discourses. Such an approach was fundamental to American structuralism, with its strict separation of levels and its claim, since espoused also by British neo-Firthians, such as M. A. K. Halliday, that the units found at one level are comprised of sequences of units found at the next lower level. Postulating that phonological segments are linearly discrete, structuralists discussed the influence of adjacent segments upon each other, such as palatalization of velars before front or of before rounded vowels, in terms of the distributional variants, or of slightly more abstract . Only in one school was the segmental discreteness postulate, as I shall call it, challenged in any serious way. This was in the earlier work of Firth and his students, who claimed that the seg­ ment could not be described as an element by itself, but only as a part 1 of some larger structure, such as the , word, or group. Firth did not hesitate to claim, for instance, that the notion of positional variant obscured the fact that the set of segments that can occur syllable-finally may be of a totally different nature from what can oc­ cur elsewhere in the syllable. An example of this is the status of the stop found after initial 5 in English, which most structuralist analyses of English have regarded as an unaspirated of the voiceless stop series. An equally plausible analysis would treat it as an allophone of the voiced stop series. Since English voiced stops are fully voiced only if found

~e Firthian notion 'group' corresponds vaguely to the structural notion 'major constituent'. 42

between voiced segments and are otherwise simply lax consonants, an analysis like this could go even further and say that the important con­ trast between the two sets of stops is not voicing, but tenseness. What are generally regarded as voiceless stops are in fact tense stops and what are regarded as voiced stops are lax consonants which become voiced between voiced segments by . The difference between word­ initial lax stops, with their very slight voicing, and post- lax stops, is certainly no greater than that between aspirated and unaspirat­ ed voiceless stops, and so the two analyses would seem equally valid. Under a Firthian analysis this problem would simply not arise, since the set of consonants found in post-sibilant position would be considered a formally distinct class of segments unrelated phonologically to the seg­ ment classes found in other canonical positions. 2 More recently, the segmental discreteness postulate has been given formal treatment and has been made the very center of phonological and phonetiC analysis in generative , where even the form of phono­ logical rules forces the linguist to think in terms of segments, rather than in terms of other, perhaps suprasegmental, units. The segmental discreteness postulate works impressively well for a large number of phonological problems, a typical case being the standard analysis of the homorganic nasal assimilations found in many languages. In Yoruba, for example, the continuous aspect marker n is prefixed to the verb stem, abutting on a consonant. The n then becomes homorganic to the consonant, as shown below: , (1) n + bQ + mbQ 'coming' n + ft + ~ft 'taking' n + ta + nta 'selling' , n + ja + Dja 'fighting' , , n + ge + Qge 'cutting'

2For a further exposition of Firth's approach to phonology see Palmer [1970]. Palmer's collection includes several important papers done within the Firthian tradition, including a paper by Robins which is an excellent summary statement of prosodic theory. 43

The rule for this assimilation can be stated quite generally as in (2) : (2) ["Mt~iorJ [-VOCaliC] + i3coronal / r~eri~]i3coronal +nasal high -- ~high ~back back 2. Non-discreteness in Accentual Phonology The possibility that there are some phonological phenomena which do not yield to a strictly segmental treatment has rarely been raised and then only with reference to so-called supra-segmental phonology. That is, with reference to the analysis of accentual systems. Consider, for example, the data presented by Woo [1969: 16-17] on Bambara accent. Woo claims that Bambara lexical items can be analyzed into two accentual classes, high-toned (3a) and low-toned (3b).

( 3) a. High-toned class b. Low-toned class , , yfd 'tree' muso 'woman' , , , mangoro 'mango' ny)m~1n 'good' tfnnan 'black'

Phrases consisting of a noun and one or more adjectives have high tone on all of the last word. All other words take the tone of the noun, the first word, on all syllables. This generalization also ap­ plies to the low tone found on the second syllable of manger6. The fol­ ].owing examples illustrate this:

( 4) manger6 nylman + mang6r6 nyfman 'good mango' muse nylman + muse nyfm~n 'good woman' , , muse tfnman nylm~1n + muse tlnman nyfman 'good black woman' yfd tfnnan nylman + yfd t fnman nyfnan 'good black tree'

The evidence from Japanese is similar. A lexical item in Japanese is specified in the lexicon as having accent on anyone mora or as being unaccented. According to Woo [1969:5-6], whose discussion is based on McCawley [1968], pitch is assigned to a word on the basis of accent. Her generalization is that "excepting the first syllable, which is low­ toned unless accented, all syllables preceding and including the accented syllable are high-toned, and all syllables following the accented 44

syllable are low toned" [Woo 1969 :6]. This is illustrated below, where accent is indicated by a double apostrophe over the of the accent­ ed syllable:

( 5) mlyako -+ mly~ka 'city' (Japanese) atamg -+ at~m~ 'head' kok8ro -+ kokaro 'heart' kgbuto -+ kabuto 'he1lllet'

In certain types of phrase, as illustrated in (6), the tone pattern does not correlate directly with lexical accent, because all accents in the phrase except the first are deleted.

(6) mlyako d~su 'It's a city.' atamg dgsu at~m~ desu 'It's a head.' atamg dgr tgru atama dattaru 'If it were a head ••• '

Thus in Japanese and Bambara there are rules which effect the entire ac­ centual pattern of the word. These rules can be written segmentally, using one of several possible conventions for indicating that all phono­ logical material except accent is to be ignored, or they can be written in terms of lexical items, deleting or neutralizing accentual features of whole morphemes. That these are not segmental, but rather morphological or string pro­ cesses is illustrated by data from Mende [Spears 1967]. Mende has six possible tone sequences or contours on the lexical item, regardless of the length of the item. Thus a falling contour can have the following realizations:

'price' too 'year' ka tl 'hoe' talama , junction'

In certain noun compounds and noun-adjective constructions, the second element loses its tone pattern entirely, becoming non-high throughout. If the first element has a final high tone, the first syllable or noun 45

of the second element assimilates to it. Final rising and falling tones on the first element are then simplified to low and high tones, respectively:

(8) b~~ nlna 'new pants' + b~l~ nlna , , , , pundrl nyamu 'ugly mosquito' + pund ( nyamu ndomaa nyandee 'beautiful shirt' + ndoma nyandee p~l~ panda 'good house' + p~14 panda

The assimilation of the first low-toned mora of the second element is a segmental process, but to describe the tone-pattern neutralization, which has syntactic rather than phonological conditioning, as a segmental process misses the atomic nature of the tone pattern. The pattern func­ tions as a unit, not as a sequence of tone-bearing segments.

3. Non-discreteness in Segmental Phonology Accentual phenomena such as those discussed above do not constitute clear difficulties for current theories of phonology.3 Discussions of segmental phonology have noted similar phenomena in vowel harmony. Lightner [1965] treats Classical Mongolian vowel harmony as a morpheme feature. Prosodic phonologists have treated such disparate phenomena as and retroflexion as fUnctions of units ranging from the syl­ lable to the word group, but in every one of these cases the long compo­ nent, to use Harris' [1944] term, is coterminus with some structural unit. A number of formalisms exist for handling features of such struc­ tural units, and. they will not be discussed here. The relevant cases are those in which segments which are a part of some formative or adja­ cent parts of distinct formatives behave in a way which requires a 'false step' analysis (Zwicky 1973]. Such a case is French vowel nasali­ zation. The standard analyses of' French nasalization require something like the following two rules [Schane 1973]:

a. V + [+nasal] / rc J ([ -]) {c } L+nas~ JI

3See Leben [1973] for extensive discussion of the representation of tone. 46

( I I = pause boundary)

b. [~nasJ ¢ / _ ([-segment]) {~I}

These rules are fUnctionally coupled in a way which has not been forma­ lized, but (9b), a supposedly independently motivated rule, applies only to the output of (9a), since all nasalized vowels in French arise from (9a). Thus (9a) creates a form which is not possible in French, a nasal­ ized vowel followed by a . (9b) is needed to correct this false step. In effect (9) represents a special type of global rule application: (9a) can apply only if (9b) also applies. What actually obtains in French is an alternation between nasalized vowels and oral vowel + nasal consonant sequences. There is no evidence either for nasal assimilation or for nasal consonant deletion. Rather, two adja­ cent segments fUse into one segment sharing some properties of both. Such an alternation cannot be captured with existing phonological ap­ paratus without a false step analysis.

3.1 Root-final palatals in Tswana. Another such case is the analysis of root-final palatals in Tswana, a Southern Bantu language. 4 Tswana forms the passive and causative verb stems by adding the suffix, or ex­ tension, -Iwa (passive) and -Isa or -va (causative) to the root, as in (10):

(10) -aga 'build' -aglwa 'be built' -b8ntsha 'show' -b8ntshlwa 'be shown' -8kMsa 'add to' -8k£dlwa 'be added to' -bua 'speak' -bulwa 'be spoken'

In many cases twu forms are possible: a form in which the fUll suffix is present, as in (10), and also a form in which the has been deleted, as in (11):

4The data for this discussion are from Cole [1955J. The problem was brought to my attention by Gerard Dalgish, a graduate student at the University of Illinois. 47

(11) -boniwa/-bonwa 'be seen' -ratlwa/-ratwa 'be loved' -roklwa/-rokwa 'be sewn' -aglwa/-agwa 'be built'

Stems which have final labials also permit the two forms. but in addi- tion, the labial becomes a palatal if the deletes:

(12) -bopa 'mould' -bopiwa/-botswa 'be moulded' -gapa 'plunder' -gaplwa/-gatswa 'be plundered' -tlhaba 'stab' -tlhablwa/-tlhaJwa 'be stabbed' -roba 'break' -r8blwa/-roJwa 'be broken' -b8fa 'bind' -b8flwa/-bMwa 'be bound' -Iefa 'pay' -Ieflwa/-Ieswa 'be paid'

The alternation in (12) could be described by an analysis which palatal­ izes labials and then deletes the I. Thus we would have derivations like the following:

bop + Iwa batS + Iwa palatalization batS + wa deletion

Such an analysis would seem to make use of the i-deletion rule which the grammar must have anyway, to account for the forms in (11). There are, however, some uncomfortable consequences associated with this sort of analysis. First of all. the palatalization rule is about as unnatural a rule as can be imagined. Not only can it not involve as­ similation. since there is no articulatory resemblance between labial consonants and palatal vowels. but in just those cases where we would most expect to find articulatory assimilation. that is. with alveolars and velars. there is no palatalization. Thus the palatalization rule is exactly the inverse of what one would expect a palatalization rule to be. palatalizing labials but not palatalizing articulatorily adjacent segments. A further problem with the analysis represented by (13) is 48

that I-deletion, normally an optional rule,5 is obligatory just if palatalization has applied. It cannot be made obligatory just af'ter labials, because as shown in (12), pi, bl, and fl sequences are all possible. The rules of palatalization and I-deletion must, then, be coupled in some way. This can be done by noting in the grammar that they are conjunctively ordered, even though they are not collapsible, or by writing a single transformational rule which would handle the two processes simultaneously, as in (14):

(14) SD: X [+Obstruej+anterior + [-obstruenj+high -coronal -back 1 2 3 4 5

SC: 1 [+Ob't~emJ-ant lor 5 +coronal +high

But (14), on the other hand, is redundant in that it incorporates the independently motivated rule of I-deletion. Having to choose between these analyses would be neither easy nor pleasant, but fortunately there are further data which render such a choice unnecessary. The causative stem is formed by adding the suffix -va or -Isa to the verb root:

(15) -tshega 'laugh' -tshaglsa 'cause to laugh' -Ioga 'knit' -Ioglsa 'cause to knit' -fena 'succumb' -fanya 'overcome' -Iekana 'be equal' -Iekanva 'make equal'

Stems ending in -f may also take either ending, but if the ending is -va, the resulting forms are as shown in (16):

5In fact, Cole [1955:192] claims that normally the deleted form is preferred. Only in two-syllable, vowel-initial roots is the I normal­ ly not deleted. 49

(16) -Ieofa 'sin' -Ieodhwa 'cause to sin,6 -tlhalefa 'become wise' -tlhaleUhwa 'make wise' -natefa 'be nice' -natedhwa 'make nice'

These forms, which have labialization in the causative although the causative suffix does not contain a labial, raise the question of what the source is of the labialization in palatalized root-final labial con­ sonants. The only possible source in (16) is the f itself, which fuses completely with the y to produce the segment tshw7 with the features [-, +coronal, +anterior, +round, +tense, +strident, -]. Such segments represent a compromise between the articu­ latory properties8 of the two segments which have fused, as illustrated below: a. p + = dw voiceless high voiceless labial palatal labialized unaspirated vowel b. ph + = tShw voiceless high voiceless labial palatal labialized aspirated vowel aspirated stop palatal affricate

6 Cole [1955:205] notes that in many cases the W seems to be disap­ pearing, but he argues that the forms with the w are definitely older than those without. 71 don't know why fy produces the affricate tshw and fl the sw. It could be due to the greater closure of y over against that of i, but this is only a guess. 81 have avoided the use of distinctive features of any variety for fairly obvious reasons. Basically they involve the facts that no cur­ rent feature framework allows labials and rounded segments or aspirated consonants and to be treated as a natural class. As the data and discussion show, both claims are necessary. To avoid having to re­ vise a feature framework on the spot, I have opted for the use of more traditional articulatory description. 50

c. b + dzw voiced high voiced labial palatal labialized stop vowel palatal affricate d. f + sw voiceless high voiceless labial palatal labialized fricative vowel e. f + Y tShw voiceless palatal voiceless labial semi-vowel aspirated fricative labialized palatal affricate

The fusion in (17 e) is what is fOlIDd in the causative forms, where, in the absence of any w, labialization occurs and can only be attributed to f maintaining its labiality even after fusion produces a palatal. The analogous fusions in (17a-d) represent the processes involved in the passive verb forms. The fused form will, of course, have the w of the wa suffix following the p + I fusion at some point in the derivation. but it cannot be claimed that the w of botswa must represent the suf­ fix. Just as strong a case exists for attributing it to the p, just as the causative fusion in natetshwa gets its w from the f. The w of wa, may then be deleted after a labialized consonant. In fact, Cole [1955:34] notes that " ••• labialization of labial consonants is not per­ missible in the sound structure of Tswana." This would block sequences of labialized consonant plus w too. (17b) and (17e) are an interesting pair in the way that aspiration is treated. In both cases the fused form is tshwa, arising in (17b) from phiwa and in (17e) from fya. The source of aspiration in (17b) is obvious. The form contrasts with an unaspirated plwa in that the latter fuses to lIDaspirated tswa, and so the aspiration must be due to the fact that phlwa is aspirated. In (17e) the only possible source for the aspiration is the fact that the lIDfused form contains an f, and the aspiration is a reflection of that. The occurrence of aspiration in fusions involving a fricative will be discussed further below. The palatalizations illustrated in (17) are also found in the diminu­ tive forms of nouns, formed by adding the suffix -ana/-ane to the noun stem. Historically this suffix began with a palatal glide y or a palatalized velar fricative yY, and the historical palatal remains evi­ dent in the alternations of labials and palatals illustrated in (18). These alternations are identical to those represented in (17), including the alternation of fy with tShw:

(18) kolob~ 'pig' kolojwane 'piglet' ts~b~ 'ear' ts~jwana 'sm. ear' kobo 'cloak' kojwana 'sm. cloak' tshipl 'iron' tshitSwana 'sm. piece of iron' mogopoA 'bowl' mogotswana 'sm. bowl' lesapo 'bone' lesatswana 'sm. bone t

ts~ph~ , springbok ' tshetShwana 'springbok kid' marophl 'blisters' marMshwana 'sm. blisters' gaufi 'nearby' gautShwane 'fairly nearby' sefMu 'blind person' sefatshwana 'It. blind person' phofu 'eland' phMshwana 'young eland'

As in the causative (16) there is no labial glide present anywhere in the environment, and so the labialization must again be attributed to the original labial obstruent. To summarize this discussion of the data from Tswana, I will stress the following points. Certain palatals arise in ways which do not sug­ gest natural assimilation rules. That is, labials palatalize, but nei­ ther coronal nor back consonants do so. Palatalization always involves the deletion of the vowel or glide which conditions it. In all cases the which results is labialized, but this labialization is not always attributable to a following w, as it is in the passive. In some cases it is the labial itself which must be posited as the source for the labialization. Finally, the resulting palatal may also 52

be aspirated if it is a non-continuant and derived from an aspirated or fricative consonant. This then suggests that the palatalization does not represent an assimilatory process, but rather the merger or fusion of two segments so that certain distinctive articulatory properties of both original segments are still present.

3.2 Voiceless nasals in Yatye. Another synchronic alternation can be found in the contraction phenomena of Yatye.9 When a CV verb is follow­ ed by a consonant initial noun or adverb, the final vowel of the verb deletes, resulting in a . If the initial consonant of the verb is a voiceless non-strident fricative and the initial consonant of the noun or adverb is a nasal, the two merge into a voiceless nasal with the same pOint of articulation as the original nasal. Otherwise the nasal assimilates in voicing to the initial consonant of the verb.

agu mOna agmOna ahy~ me akpa masl akp~6osl a~u many)

ahy~ n~ an~o The first two examples in (19) suggest that Yatye has an independently motivated rule of nasal voicing assimilation. The data would then lend themselves to an analysis in which voicing assimilation would be follow­ ed by consonant deletion. However, there is no independent evidence of such consonant deletion, and that rule would apply only to the output of voicing assimilation. The last form in (19) disconfirms such a solution,

since the labiality of ~ is preserved as labialization of the fused segment. The labialization cannot be derived from the u, since labi­ alization would then be expected in agmona as well.

3.3 Vowel coalescence in QwQn Afa. qw~n Afa is a dialect of Yoruba in which verb-nominal sequences contract according to regular phonologi­ cal rules [Awobuluyi 1972], unlike the highly irregular contractions of Standard Yoruba. In some cases the final vowel of the verb deletes, and

9Yatye is a Kwa language spoken by about 20,000 people in the Benue­ Plateau and Southeast States of Nigeria. The data are from my own field work in 1965-1967. 53

in others the vowel of the verb assimilates to the initial vowel of the noun. The cases relevant to the problem of fusion are those in which the final vowel of the verb is non-high, e, e, a, 0, or 0, and the initial vowel of the noun is non-covered, I, e, 0 or u. The fol­ lowing fusions occur:

(20 ) a. a + + e da lwe + dewe 'buy book' b. a + e + e da ehwe + dehwe 'buy house'

c. a + u + 0 d'a 'J'u u + doJu 'buy pounded yam' d. a + 0 + 0 da 6pu + d"oopu ,10 'buy dog'

e. e + u + 0 te uJu + toJu ' st ep on pounded yam' f. e + 0 + 0 te 6pu + t"oopu ,10 'step on dog' g. 0+1 + e do lwe + dewe 'burn books' h. o + e + e do ehwe + dehwe 'burn houses i. e + u + 0 be uJu + boJu 'look for pounded yam' j. 0+1 + e , go lwe + gewe 'look at book'

These fusions can be captured by the following transformational rule:

(2l) V -high -low I~high l + Lacovere~ r:~~li3back acovered yround Sback L yround

~O"::-le,,"C'\-!(;_:_ length is ::'r-Oill ·~he L11 tone sequence ::md =.s:ot

~ -L ~ or _ 3~~_J:i.li").~J:: A.!",3 s :~m:::.J..s..t ion" :\'".,.rot·lll uyi

That is, the segment resulting from the fusion is a with the coveredness of the first vowel and the backness and roundness of the sec­ ond vowel. It is a compromise between the two original vowels. Assimilatory alternatives to (21) are possible. For example one could assimilate the second vowel in tongue height and coveredness to the first vowel and then delete the first vowel.

( 22) a. -+ [~covereJ [-highacovered J / [v-high J # acovered b. [~higJ -+ ~ / # [~highJ-low

This analysis seems well-motivated, since Awobuluyi gives numerous other forms in which assimilation occurs, and there are also cases where a vowel is deleted. However, all other assimilations are complete, not partial, and all; but a well-defined subset of cases involve the assimi­ lation of the first vowel to the second, rather than as in (22a). If (22a) were replaced with a rule which assimilated the first vowel to the second, in keeping with the general nature of vowel assimilation in Qw9n Afa, it would then be necessary to rewrite (22b) to delete the second

vowel. This would be unnatural in Qw~n Afa, since under no other condi­ tions does the second vowel delete. Thus neither segmental treatment of the data is without serious difficulties. The fusion analysis is clear­ ly superior since it avoids these difficulties.

4. Diachronic Evidence for Fusion The Tswana and Yatye data presented above show cases of active syn­ chronic alternations which can best be described as segmental mergers or fusions. There are also examples of diachronic fusions, typical ex­ amples of which may be the fate of the PIE labio-velars in Greek, and in of the PIE aspirates.

(23) PIE Greek Latin *kw hippos equus 'horse' poteros quod 'which' 55

*gw balno venlo 'come'

bos 'cow' cf. Skt. gau~ *bh phrator frater 'brother' phero fero 'bear' *dh tithemi facio 'do' eruthros rufus 'red' *gwh formus 'warm' cf. Pers. garm phonos defendo 'strike' cf. Skt. hantl

A case of diachronic fusion which is directly analogous to the Tswana data is the correspondence sets involving labialized palatals in several Yatye dialects.ll Where Alifokpa has the labialized palatals cw, jw, and njw, Akweya has strengthened the labialization to f or v and Ijiegu has changed the labialized palatals to the palatalized py, by, and mby.

(24) AlifokJ2a Akwel.a IJie~ (cwende fcfend~ (pyendE 'pot' , , ecwu ecfu epyu 'head ' JWu jvu byu 'drink' I njw( I nJv( Imby( , germinat e '

5. Evidence From Child Language Acquisition A third type of evidence that speakers can treat segment sequences as units comes from child language acquisition. The subject was Aaron, age 36 months. At the time he was producing some unusual substitutes for initial s + nasal clusters. After going through the stages many children seem to pass through as illustrated below in (25), Aaron then entered a stage represented in (26).

l~he Alifokpa and Ijiegu dialects are separated only by a river, and they are mutually intelligible. Akweya is separated from them by about sixty miles and a hundred and seventy years, dating from the Fulani jihad of 1804. 56

(25) 30 months 34 months mok rtmok 'smoke' mayo rvnayo 'smile' meant II rumantll 'Samantha' nek gnek 'snake no nno 0 'snow' nupl gnupl 'Snoopy'

(26) 36 months phOk 'smoke' h ___ p ayo 'smile' h_ t p an II 'Samantha' thek 'snake' tho 'snow' t h-Upl . 'Snoopy'

By the age of approximately 38 months Aaron has mastered the 5 + nasal cluster, but what is interesting is what he was doing up to that time. At 30 months the sm and sn clusters were produced simply as nasals. At 34 months the initial voiceless fricative began to show up as a devoicing of the first part of the nasal. At 36 months he had, in effect, mastered the cluster, but what he did, as illustrated in (26), was merge or fuse the two consonants into a single unit showing the distinctive articulatory properties of both the s and the nasal. h (27) a. s + m p voiceless labial aspirated fricative nasal voiceless labial stop b. s + n t h voiceless alveolar voiceless fricative nasal aspirated

The fusion combines the oral closure of the nasal with the and obstruency of the s to produce a stop homorganic with the original 57

nasal. Just as in Tswana, the original fricative shows up as aspiration in the fusion. The nasality, as shown in (26) is moved over to the vow­ . The important point is that all of the essential phonetic informa­ tion is present in the 36 month forms. It has simply been restructured and redistributed. A similar merger cOllDllon in child language is the fusion of initial tr to f or fw. Phonetically English tr becomes an aspirated labialized prepalatal affricate. The fw sequence treats the aspiration and friction of the original cluster as an f, retaining the original labialization.12

6. The Inadequacy of a Rule Ordering Solution There appear to be certain phonological processes in natural lan­ guage which cannot be stated by segmental rules without serious formal difficulties. Data of the type discussed above strongly suggest that phonological theory requires a formal notion of string operation analo­ gous to grammatical transformations, in addition to the well-developed formalism for segmental rules. Such a formal addition defines a class of phenomena in which two adjacent segments are replaced by a single segment sharing some features of each. If described segmentally, two rules are required. One of these defines an assimilation of one segment to the other if and only if the second rule deletes the conditioning segment after the assimilation rule has applied. The synchronic alter­ nations in Tswana, Yatye, French, and Qw9n Afa illustrate this. In Tswana, the labial palatalization rUle can apply only if the or y is doomed to deletion. l-deletion is optional otherwise, and there is no independent motivation for y-deletion. Thus the disappearance of the and y after palatalization is due to some other factor than l-deletion. Even if or y deletion could be motivated here, the problem would exist that palatalization could apply only if the deletion rule would follow it. That is, palatalization would be a global, look­ ahead rule. This analysis, however, fails on independent grounds since it cannot account for the labialization which is found after

l~his fusion was the source of occasional consternation when Aaron would ask his nursery teacher for a truck. 58

palatalization. There simply is no source for the w, unless the labi­ alized palatal is regarded as the fusion of a labial and a palatal. In the Yatye data, exactly the same formal situation obtains. If a ­ less non-strident fricative abuts on a nasal, the result is a voiceless nasal. Segmentally this could be handled by first devoicing the nasal and then deleting the fricative, But there is no independent motivation for the fricative deletion rule, and the devoicing rule would have to look ahead to see that the fricative does delete. Once again, the seg­ mental analysis, which requires a forward looking global ordering, is inadequate because it fails to account for the labialization of forms like awwaanyl in (19). This property can be traced directly back to the bilabial fricative in the string which fused to produce the voice­ less nasal, but it cannot be explained under the segmental discreteness postulate. French nasalization has been a problem for generative phonologists precisely because of the feeding relation between nasalization and nasal deletion, and because of the relative simplicity of the data it has been difficult to motivate one or another analysis. The segmental fusion phenomena I have described show that for certain purposes natural lan­ guage may treat segment sequences as units. The cases susceptible to a fusion analysis contain as a subset cases which can also take a particu­ lar type of global analysis, namely that involving look-ahead rules, but the fusion phenomenon may also involve changes, such as the labializa­ tion of the derived segment in Tswana and Yatye, which are not naturally explainable by the global analysis. Therefore if the fusion phenomenon is well motivated, then this class of look-ahead global rules must be disallowed by the metatheory. By restricting the metatheory in this way it becomes possible to choose one of the possible analyses for problems such as French nasalization as preferred, namely, the fusion analysis.

7. On the Nature of the FUSed Segment One important question remains to be dealt with: the difficult mat­ ter of predicting what features the fusion will retain from the original segments. The cases which have been examined suggest several possible 59

generalizations. In French a vowel and a nasal fuse to a nasalized vow­ el. That is, the vowel takes on what might be considered, for reasons that are difficult to state formally, the most prominent feature of the consonant--its nasality. Changes in vowel quality in French defy ex­ planation, but the fused segment is in an obvious way a compromise be­ tween the original segments, exhibiting the major class feature of the vowel as well as the most striking feature of the consonant. In a vari­ ety of other languages it can be shown that vowel nasalization arises from a NY segment rather than from a VN sequence, as in French. Hyman [1972] argues this convincingly for Gwari and Williamson [1973] demon­ strates it for Ijq. The obvious prediction is that a vowel and a nasal will fuse to a nasalized vowel, regardless of the original order of segments. In the Qwqn Afa case, the original segments are both vowels. That is, they share the same major class features. The fused segment is again a compromise, but the type of compromise is clearer than in the case of nasalized vowels, as the following charts show.

(28) a. i u b. e 0 1 1 1 1 E ::> /' /" a " a c. u d. rJ (~ E ) ::> e f

e. U f. ! ! e • 0 e~ 0

Whether the fusion is described in terms of features, as in (21) or in terms of tongue position, as in the charts above, the fused segment is always that 'Towel in the seven vowel system of Qw'l'n Afa that most nearly represents the mid point between the two original vowels. It is clear, 60

then, why sequences that differ by only one position do not fuse. and u have no Compromise vowel in the system. QwQn Afa has no u or

+. Similarly a can fuse with neither £ nor 0, since it is adjacent to each. There must be at least two significant differences between the vowels for fusion to be possible. In a sense the voiceless nasals in Yatye are a similar case. Both original segments are consonants. That is, they are both [+consonant, -vocalic]. While they differ as to [continuant] and [], they both allow an uninterrupted flow of air through the vocal tract. The compromise segment is also [+consonant, -vocalic] and shares the [-voiced] of the fricative and the point of articulation of the nasal. The child language data are similar, in this case the voicelessness and

velic closure of the 5 changing the nasal into the corresponding voice­

less stop. The continuantness of the 5 is preserved in the aspiration of the stop, although a feature representation would not show this clearly. Two interesting cases are Tswana and Indo-European. In these cases the original segments differ radically. A and a high or glide are about as distinct from each other as two seg­ ments can be, but the fused form in Tswana is still a clear compromise sharing the voicing and major class features of the consonant and the tongue position of the vowel. The Indo-European data are similar. If the labial-velars can be interpreted as segment sequences then kw + p represents a fusion nearly identical to that in Tswana palatals. The Latin f, coming from various aspirates in PIE shows the reverse of the

sort of fusion Aaron employed in treating the continuantness of 5 as aspiration. Instead, the aspiration shows up as a voiceless continuant. A general fusion schema apparently cannot be stated in terms of bi­ nary features. Rather, a scale of sonority is needed. At the one end of this scale are stop consonants and at the other are vowels. A partial scale would be something like the following,13 using only the segment types exemplified in the data above:

13Scales of this sort have been proposed elsewhere in the literature, for example, Sadock [no date], Zwicky [1972], and Foley [1969]. (29) stops fricatives nasals glides vowels

A fUsion of two members of different classes on this scale moves in the direction of greater or less sonority, depending on which segment is closer to one of the poles. This will account for VN and NV sequences becoming nasalized vowels, since vowels are at a pole and nasals are not. It will also account for stop plus glide fusing to a stop, but it will not account for stop plus vowel fusing to a stop or for voiceless frica­ tive plus nasal fUsing to a voiceless nasal or a voiceless stop, al­ though both are reasonable compromises. Other factors seem to be relevant. First of all, voicelessness is a dominant feature. If one of the two segments is voiceless, the fUsion is invariably voiceless. The absence of the voiceless nasal in Aaron's speech may be due to the highly marked nature of this segment in child language, as compared with the more natural voiced stop. Point of ar­ ticulation is also sensitive to the sonority scale. In all cases the surviving point of articulation in the fUsion is that of the more sono­ rous member of the pair. This generalization does not, however, account for all of the sources of Latin f. The cases in which Latin f arises from a PIE labial-velar are predicted, the f taking the point of ar­ ticulation of the labial component. In all other cases it seems to be the aspiration which is maintained in the voiceless f. If one can generalize from the phonetic property of aspiration to a property of voiceless friction, then it appears that voiceless friction will be con­ sistently maintained. The alternations reported above between aspirates and fricatives suggest that voiceless friction is a linguistically sig­ nificant phonetic property, since aspirates and fricatives do appear to behave as a natural class. Interestingly enough, the one PIE aspirate which does not become f in Latin is *gwh, which becomes h.

Postscript After the revised draft of this paper was finished, Hankamer and Aissen [1974] came to my attention. To account for certain alternations 62

in , they propose a scale or hierarchy of sonority as follows: stops 5 nasals v y r vowels Examination of Hungarian data leads them to suggest that fricatives gener­

ally behave like 5 and that the ranking of glides and liquids relative to each other is language-specific. Among the phenomena which they sug­ gest are influenced by the are syllable structure, syllabicity, the determination of natural classes, phonological change, variability, and assimilation. On the basis of the data presented in this paper, fUsion must be added to that list.

REFERENCES

Awobuluyi, Qladele. 1972. The Morphophonemics of Qw~m Afa. Ibadan Research Notes 5, Nos. 2 and 3, pp. 25-44. Cole, Desmond. 1955. An Introduction to Tswana Grammar. London: Longmans, Green, and Co. Foley, James. 1969. Morphophonological Investigations II. Paper read at the UCLA Conference on Historical Linguistics in the Perspective of Transformational Theory. Hankamer, Jorge, and Judith Aissen. 1974. "The sonority hierarchy." Papers from the Parasession on Natural Phonology. Chicago Linguistics Society. Harris, Zellig S. 1944. "Simultaneous components in phonology." Language 20: 181-205. Hyman, Larry M. 1972. "Nasals and nasalization in Kwa." Studies in African Linguistics 3:167-206. Leben, William R. 1973. "The role of tone in segmental phonology." In Larry M. Hyman, ed., Consonant Types and Tone, Southern California Occasional Papers in Linguistics, No.1. Lightner, Theodore M. 1965. "On the description of vowel and consonant harmony." Word 19:376-387. McCawley, James. 1968. The Phonological Component of a Generative Grammar of Japanese. The Hague: Mouton and Co. Palmer, F. 1970. Prosodic Analysis. London: Oxford University Press. Sadock, Jerrold. (no date). Non-discrete (Indiscreet?) Phonology. Unpublished ms. Schane, Sanford. 1973. "The treatment of phonological exceptions: the evidence from French." In Braj B. Kachru, et al., eds., Issues in Linguistics: Papers in Honor of Henry and Ren~e Kahane, 822-

S?,~. TTY"h~;·"":· -·y-,..';"':TC"':>.,...,.....··- 63

Spears, Richard. 1967. An IntroductorY Course in Mende: Evanston: African Studies Center, Northwestern University. Williamson, Kay. 1973. "More on nasals and nasalization in Kwa." Studies in African Linguistics 4:115-138. Woo, Nancy. 1969. and Phonology. Unpublished mimeo. Bloomington: Indiana University Linguistics Club. Zwicky, Arnold M. 1972. "Note on a phonological hierarchy in English." In Robert p. Stockwell and Ronald K. S. Macaulay, ed., Linguistic Change and Generative Theory. Bloomington: Indiana University Press. Zwicky, Arnold M. 1973. "Taking a false step." Working Papers in Linguistics 14:100-112. Columbus: The Ohio State University. 64